locale: symlink identical langpack LC_CTYPEs to the C.utf8 version

Almost half the LC_CTYPE files in langpacks are identical to the C.utf8
variant which is installed by default.  This commit converts these to
symbolic links instead, saving ~350K of installed size per langpack.

Rawhide commit: a0bdca4f3f

Also:

 * locale: make install-locale-files instead of install-locales
   - Drop glibc-fedora-localedef.patch
   - Rawhide commit: cf979ff812

 * localedef: Handle symbolic links when generating locale-archive
   - Add glibc-localedef-handle-symlinks.patch
   - Upstream commit: ea89d5bbd9e5e514b606045d909e6ab87d851c88

glibc-fedora-localedef.patch

@@ -1,21 +0,0 @@
-Short description: Fedora-specific glibc install locale changes.
-Author(s):  Fedora glibc team <glibc@lists.fedoraproject.org>
-Origin: PATCH
-Upstream status: not-needed
-
-The Fedora glibc build and install does not need the normal install
-behaviour which updates the locale archive. The Fedora install phase
-in the spec file of the rpm will handle this manually.
-
-diff --git a/localedata/Makefile b/localedata/Makefile
-index 0eea396ad86da956..54caabda33728207 100644
---- a/localedata/Makefile
-+++ b/localedata/Makefile
-@@ -413,6 +413,7 @@ define build-one-locale
- 	echo -n '...'; \
- 	input=`echo $$locale | sed 's/\([^.]*\)[^@]*\(.*\)/\1\2/'`; \
- 	$(LOCALEDEF) $$flags --alias-file=../intl/locale.alias \
-+		     --no-archive \
- 		     -i locales/$$input -f charmaps/$$charset \
- 		     $(addprefix --prefix=,$(install_root)) $$locale \
- 	&& echo ' done';

glibc-localedef-handle-symlinks.patch

@@ -0,0 +1,27 @@
+commit ea89d5bbd9e5e514b606045d909e6ab87d851c88
+Author: Arjun Shankar <arjun@redhat.com>
+Date:   Thu Feb 24 21:43:09 2022 +0100
+
+    localedef: Handle symbolic links when generating locale-archive
+    
+    Whenever locale data for any locale included symbolic links, localedef
+    would throw the error "incomplete set of locale files" and exclude it
+    from the generated locale archive.  This commit fixes that.
+    
+    Co-authored-by: Florian Weimer <fweimer@redhat.com>
+    
+    Reviewed-by: Carlos O'Donell <carlos@redhat.com>
+
+diff --git a/locale/programs/locarchive.c b/locale/programs/locarchive.c
+index 45408c26c169ff05..eeb2fa6ffecd1dba 100644
+--- a/locale/programs/locarchive.c
++++ b/locale/programs/locarchive.c
+@@ -1397,7 +1397,7 @@ add_locales_to_archive (size_t nlist, char *list[], bool replace)
+ 		    {
+ 		      char fullname[fnamelen + 2 * strlen (d->d_name) + 7];
+ 
+-		      if (d_type == DT_UNKNOWN)
++		      if (d_type == DT_UNKNOWN || d_type == DT_LNK)
+ 			{
+ 			  strcpy (stpcpy (stpcpy (fullname, fname), "/"),
+ 				  d->d_name);

glibc.spec

@@ -152,7 +152,7 @@ end \
 Summary: The GNU libc libraries
 Name: glibc
 Version: %{glibcversion}
-Release: 3%{?dist}
+Release: 4%{?dist}
 
 # In general, GPLv2+ is used by programs, LGPLv2+ is used for
 # libraries.
@@ -217,7 +217,6 @@ rpm.define("__debug_install_post bash " .. wrapper
 # - For new patches follow template.patch format.
 ##############################################################################
 Patch4: glibc-fedora-linux-tcsetattr.patch
-Patch6: glibc-fedora-localedef.patch
 Patch8: glibc-fedora-manual-dircategory.patch
 Patch9: glibc-rh827510.patch
 Patch13: glibc-fedora-localedata-rh61908.patch
@@ -225,6 +224,7 @@ Patch15: glibc-rh1070416.patch
 Patch17: glibc-cs-path.patch
 Patch23: glibc-python3.patch
 Patch30: glibc-deprecated-selinux-makedb.patch
+Patch31: glibc-localedef-handle-symlinks.patch
 
 # Temporary until gcc is patched to have -mlong-double-64 override
 # -mabi=ibmlongdouble instead of conflict with it.
@@ -1218,11 +1218,11 @@ done
 pushd build-%{target}
 %make_build install_root=%{glibc_sysroot} install
 %make_build install_root=%{glibc_sysroot} \
-	install-locales -C ../localedata objdir=`pwd`
+	install-locale-files -C ../localedata objdir=`pwd`
 popd
-# Locale creation may produce different groups of hardlinks in an
-# unpredictable manner.  Re-grouping makes those differences go away.
-hardlink %{glibc_sysroot}/usr/lib/locale
+# Locale creation via install-locale-files does not group identical files
+# via hardlinks, so we must group them ourselves.
+hardlink -c %{glibc_sysroot}/usr/lib/locale
 
 # install_different:
 #	Install all core libraries into DESTDIR/SUBDIR. Either the file is
@@ -1344,6 +1344,22 @@ $olddir/build-%{target}/elf/ld.so \
 # glibc-all-langpacks versions)
 ln locale-archive locale-archive.real
 
+# Almost half the LC_CTYPE files in langpacks are identical to the C.utf8
+# variant which is installed by default.  When we keep them as hardlinks,
+# each langpack ends up retaining a copy.  If we convert these to symbolic
+# links instead, we save ~350K each when they get installed that way.
+#
+# LC_MEASUREMENT and LC_PAPER also have several duplicates but we don't
+# bother with these because they are only ~30 bytes each.
+#
+# Note: This must be done AFTER locale-archive generation since the locale
+# archive is incorrectly generated when symbolic links are present.
+pushd %{glibc_sysroot}/usr/lib/locale
+for f in $(find eo *_* -samefile C.utf8/LC_CTYPE); do
+  rm $f && ln -s '../C.utf8/LC_CTYPE' $f
+done
+popd
+
 # Create the file lists for the language specific sub-packages:
 for i in eo *_*
 do
@@ -2161,6 +2177,13 @@ update_gconv_modules_cache ()
 %files -f compat-libpthread-nonshared.filelist -n compat-libpthread-nonshared
 
 %changelog
+* Wed Mar 02 2022 Arjun Shankar <arjun@redhat.com> - 2.35-4
+- Reduce installed size of some langpacks by de-duplicating LC_CTYPE.
+- Add glibc-localedef-handle-symlinks.patch so localedef can handle symbolic
+  links when generating locale-archive.
+- Drop glibc-fedora-localedef.patch and adjust locale installation
+  accordingly so that installed content remains unchanged.
+
 * Mon Feb 28 2022 Florian Weimer <fweimer@redhat.com> - 2.35-3
 - Auto-sync with upstream branch release/2.35/master,
   commit 732dd3a63d39c7ca77e817b462285c14551c8b49: